The world’s most powerful tech companies are racing to build enough data center infrastructure to support their artificial intelligence ambitions — and many are having to get creative with how they supply power.
As the power crunch intensifies, a handful of solutions designed to skirt the traditional cost-sharing model, and the utility wait times, are gaining traction.
But some of those solutions, including colocation, peak shaving, and special contracts, could still end up shifting costs to consumers, said Ari Peskoe, a researcher at Harvard Law School’s Environmental and Energy Program, who recently co-authored a paper on the topic.
Colocation
In the case of colocation, power plant owners have identified the chance to make data centers their direct customers, by connecting them directly to a plant, behind the meter. Utilities have pushed back on that arrangement, Peskoe said, because it denies them the chance to build new infrastructure that would otherwise be required to connect those data centers — and increase their rate base in the process.
“The colocation debate is really a competition between the generators and utilities,” Peskoe explained. “Everybody sees that these data centers are a huge opportunity for profits, and they want to get a piece of it,”even if it means hiking prices for their smaller customers.
The cost shift Peskoe sees in the case of colocating data centers with, for example, nuclear plants is that potential generators of clean firm power may decide not to participate in the market because they can make more money by working directly with tech companies.
“All else being equal, supply goes down, costs are going to go up, at least in the short term,” Peskoe explained. “That’s the potential cost shift here is these plants basically exiting the market to have private agreements with Google, Amazon, et cetera, and that’s going to cause market prices to go even higher than they already have been.”
Take the case of Constellation Energy and Exelon, which are currently engaged in a dispute over colocating major loads with nuclear facilities.
Constellation, which owns nuclear power plants in PJM, is advocating for the Federal Energy Regulatory Commission to clarify rules governing colocated projects. Because Exelon is not delivering power to the data center, Constellation said, the data center is not a utility customer and shouldn’t have to pay any regulated delivery charges.
Exelon, meanwhile, argues that even colocated projects still benefit from the transmission system and should therefore help to cover their share of those benefits. Another potential issue, the utility said, is that colocated data centers could impact overall system reliability.
PJM’s independent market monitor, for its part, has warned that widespread colocation could have “extreme” impacts on electricity markets and prices. “Energy prices would increase significantly as low-cost nuclear energy is displaced by higher cost energy,” the monitor said.
One more obvious issue with the looming data center buildout, Peskoe said, is that ratepayers could ultimately be left paying for transmission lines built to serve data center load that never fully materialized.
He pointed to the case of AEP Ohio, which proposed a new tariff that would require data centers to sign long-term utility contracts agreeing to pay 90% of costs for maximum demand for 10 years before receiving service, including FERC transmission costs.
But while that setup at least partially shields other AEP ratepayers from data center transmission costs, ratepayers of neighboring utilities still could end up footing the bill, Peskoe said. That’s because PJM splits the costs of regional transmission lines among member utilities. And if AEP’s data center projections don’t come to fruition, or if customers cancel their projects, total regional costs could rise.
Peaking shaving and special contracts
Another solution gaining popularity is peak shaving. Like colocation, Peskoe said the approach could also cause ratepayer rates to rise.
Data centers pay both per-unit energy prices and demand charges based on their consumption during peak power demand periods. The latter charges are designed to cover the cost of building and maintaining capacity for extreme weather events: capacity that goes unused the majority of the time.
If a data center or another large industrial power consumer can predict when peak demand will happen in a given year, and use its own backup power during those hours it could substantially reduce its bill. In doing so, that customer would be shifting the cost of that extra capacity to ratepayers who don’t have similar forecast or backup capabilities.
“There’s this whole little consultancy industry of trying to predict when these [peak] hours are going to be, and therefore you can potentially turn off your grid-delivered power, turn on your backup generators…and potentially avoid these massive demand charges, which pushes them off to everybody else,” Peskoe said.
Public Utility Commissions, he added, can prevent such cost shifts by structuring rates appropriately. However, transparency into data center power contracts is often lacking.
Peskoe and his co-author, Eliza Martin, reviewed 40 state PUC proceedings about so-called “special contracts” between utilities and data centers. That review, Peskoe explained, found that many such proceedings are shielded from public view, and the only participant is the utility itself.
“The easiest thing for the utility commission to do is just to approve the agreement, particularly when there may be political pressure to bring this [data center] project to the state,” Peskoe said.
In that scenario, a special contract could ultimately shift costs to other ratepayers if a data center customer is paying the utility less than what it costs the utility to serve that customer. Utilities will then use subsequent rate cases to raise rates for other customers to cover the gap.
“That was one thing that was somewhat surprising to us, is how many of these special contracts we saw out there,” Peskoe said.
Potential solutions
There are several strategies that regulators, lawmakers, and utilities could implement in order to support data center growth without unintentionally passing on costs to consumers, Peskoe said.
Some could be done without implementing new legislation at all. For instance, Peskoe said PUCs could stop approving special or secret contracts. Instead of contracts, they could require utilities to update their tariffs for new large loads — which he anticipates would be almost entirely data centers.
“We think that’s a more procedurally appropriate way to bring these giant new customers online,” he said. “Tariff proceedings are public. Everybody kind of has a stake in it. There’s more participation. It allows for changes down the line,” Peskoe said.” Most states, he added, could probably take both of those steps.
One revealing case study, he said, again involves AEP Ohio, which recently filed a new tariff.
“The problem with the proceeding is that AEP says it has already signed up five gigawatts of data centers under the previous regime, and when they file this new tariff they say over and over again, the reason we need to file is to protect our other ratepayers,” Peskoe explained. As far as Peskoe is concerned, that’s an admission that ratepayers weren’t protected from the costs of the five gigawatts of data centers already contracted.
But there are also more comprehensive solutions that could happen via legislative changes. That’s particularly important for vertically integrated utilities, Pekoe said, pointing to Louisville Gas and Electric, which in early March submitted a proposal with the Kentucky Public Service Commission seeking permission to build a $3.7 billion gas plant with battery storage, primarily needed for data centers.
In that case, the utility hasn’t said yet how they’re going to recover costs; but Peskoe said if they go to a rate case and try to spread it out via their traditional methods, it’s going to be a “massive subsidy to that data center from everybody else.” His solution is to allow, or even require, data centers to procure their own generation.
“Make the deal just between the generator and the data center, and they still have to go through the utility to connect to the system,” he said. “So it’s not colocating…we’re just saying to and procure your own generation.”
Even more protective of consumers, he added, would be to take data centers entirely off-grid, building “energy parks” that don’t interact with the utility system at all. Such parks, though, would likely require changes in state law.
As broader, more sweeping recommendations, Peskoe argues for more transparency from utilities, and requirements for flexible operations for data centers.
Utility disclosure of data center forecasts would prevent double-counting when data centers are negotiating with multiple utilities, and would enable competitors to develop solutions that don’t rely on utility infrastructure, he said. And reducing data center power consumption during peak demand periods from the outset, and classifying them as interruptible customers in utility tariffs, would mean the system could be designed for a lower peak that accounts for data centers’ ability to self-power, he added. This is an option that a group of utilities have been exploring via the DCFlex coalition, which launched in October 2024 and recently expanded to Europe.
The bottom line, Peskoe added, is that if you can connect a data center without building more infrastructure, it could be a good thing for consumers.
